Connect

8 clients · secrets stay placeholders

JSON (mcpServers)Claude CodeCodexOpenCodeCursorVS CodeGemini CLIGoose

{
  "mcpServers": {
    "funkel": {
      "type": "http",
      "url": "https://api.funkel.ai/mcp"
    }
  }
}

claude mcp add --transport http funkel https://api.funkel.ai/mcp

[mcp_servers.funkel]
url = "https://api.funkel.ai/mcp"

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "funkel": {
      "type": "remote",
      "url": "https://api.funkel.ai/mcp",
      "enabled": true
    }
  }
}

{
  "mcpServers": {
    "funkel": {
      "type": "http",
      "url": "https://api.funkel.ai/mcp"
    }
  },
  "deeplink": "cursor://anysphere.cursor-deeplink/mcp/install?name=funkel&config=eyJ0eXBlIjoiaHR0cCIsInVybCI6Imh0dHBzOi8vYXBpLmZ1bmtlbC5haS9tY3AifQ%3D%3D"
}

{
  "servers": {
    "funkel": {
      "type": "http",
      "url": "https://api.funkel.ai/mcp"
    }
  }
}

{
  "mcpServers": {
    "funkel": {
      "httpUrl": "https://api.funkel.ai/mcp"
    }
  }
}

extensions:
  "funkel":
    enabled: true
    name: "funkel"
    type: streamable_http
    uri: "https://api.funkel.ai/mcp"
    timeout: 300

Quality 77 / 100new

protocol75 · weight 25

warn
protocol.modernnewest supported version is 2025-06-18; 2026-07-28 not supported, older versions are deprecated until 2027-07-28
n/a
protocol.statelessonly applies to 2026-07-28 servers
pass
protocol.transportstreamable HTTP
n/a
protocol.list_ttlonly applies to 2026-07-28 servers

authorizationn/a · weight 15

n/a
auth.prm, auth.as_metadata, auth.cimdno remote uses OAuth
n/a
auth.secret_in_urlno templated URL

tool hygiene79 · weight 30

pass
tools.descriptionsevery tool has a description
warn
tools.description_length1 tools have descriptions over 1,024 characters, which crowds the context
pass
tools.schemasevery tool has an object input schema
pass
tools.annotations51 of 51 tools declare readOnlyHint or destructiveHint
pass
tools.directory_hintsevery tool declares readOnlyHint, destructiveHint, idempotentHint and openWorldHint
warn
tools.token_costabout 12,029 tokens to load every tool
warn
tools.api_dump51 tools, 69% named like REST operations: looks like an API mapped one endpoint per tool

stabilityn/a · weight 15

n/a
stability.changes, stability.rug_pulltool history recorded for 3 of the 7 days needed

dependenciesn/a · weight 15

n/a
deps.known_vulns, deps.mcp_sdk_version, deps.resolvableno npm or PyPI package

51 tools, about 12,029 tokens to load them all · tool set d8cd40f005e7 · tools last changed 2026-10-17 · checked 2026-10-20 13:49 UTC

Tool changes

2026-10-17create_agentchangeddescription
before
Create a new draft agent under the request's scoped app. Two-phase: first call previews every setting; second call after user confirmation creates it. The agent's ICP starts from the app's saved ICP; icp_filters keys you supply override only those keys, every other saved field (job_titles, mandatory_keywords, excluded_companies, company_sizes, ...) is kept as-is. Do not restate the full ICP from memory for a vague request like "set it up" — omit icp_filters entirely to inherit the saved ICP untouched, and only pass the specific keys the user actually asked to change. To deliberately clear a list field, pass it as an empty array. excluded_companies stores exact current employer names. If supplied a LinkedIn company URL, Operator saves its slug name there and retains the URL separately for competitor signals; ask the user to verify the name in the approval. Supply signals only when the user selects exact agent discovery signals; each signal must include its supported platform, category, key, label, enabled state, and config. An explicit empty signals array creates no signals. Without signals, defaults seed from the merged ICP filters. A requested keyword such as product launch can be a keyword_tracking_product_launch signal with config.keyword='product launch'. Ask for missing signal details; never claim a requested setting was saved if this schema cannot represent it.
after · 93% of words in common
Create a new draft agent under the request's scoped app. Two-phase: first call previews every setting; second call after user confirmation creates it. The agent's ICP starts from the app's saved ICP; icp_filters keys you supply override only those keys, every other saved field (job_titles, mandatory_keywords, excluded_companies, company_sizes, ...) is kept as-is. Do not restate the full ICP from memory for a vague request like "set it up" — omit icp_filters entirely to inherit the saved ICP untouched, and only pass the specific keys the user actually asked to change. To deliberately clear a list field, pass it as an empty array. excluded_companies stores exact current employer names. If supplied a LinkedIn company URL, Operator saves its slug name there and retains the URL separately for competitor signals; ask the user to verify the name in the approval. Supply signals only when the user selects exact agent discovery signals; each signal must include its supported platform, category, key, label, enabled state, and config. An explicit empty signals array creates no signals. Without signals, defaults seed from the merged ICP filters. A requested keyword such as product launch can be a keyword_tracking_product_launch signal with config.keyword='product launch'. Ask for missing signal details; never claim a requested setting was saved if this schema cannot represent it. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17update_icpchangeddescription
before
Apply an ICP edit (description and/or filters JSON) to one agent. Two-phase: first call returns a confirmation token + preview; the user reviews the diff inline and confirms; second call (with confirmation_token) commits. ALWAYS call draft_icp_edit first so the user sees a before/after; never call update_icp directly without staging a diff. Pre-populate agent_name in the schema args so the preview is human-readable.
after · 83% of words in common
Apply an ICP edit (description and/or filters JSON) to one agent. Two-phase: first call returns a confirmation token + preview; the user reviews the diff inline and confirms; second call (with confirmation_token) commits. ALWAYS call draft_icp_edit first so the user sees a before/after; never call update_icp directly without staging a diff. Pre-populate agent_name in the schema args so the preview is human-readable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17approve_leadschangeddescription
before
Approve or reject one or more discovered leads. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms; second call (with confirmation_token) commits the status flip. action='approve' requires campaign_id (the campaign the leads will be associated with); action='reject' does not. The tool stamps each lead's status + reviewed_at + campaign_id; downstream campaign-engine pipelines ingest approved leads through their normal path.
after · 84% of words in common
Approve or reject one or more discovered leads. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms; second call (with confirmation_token) commits the status flip. action='approve' requires campaign_id (the campaign the leads will be associated with); action='reject' does not. The tool stamps each lead's status + reviewed_at + campaign_id; downstream campaign-engine pipelines ingest approved leads through their normal path. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17pause_agentchangeddescription
before
Pause an active agent so it stops discovering new leads. The agent's existing leads remain in the list. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms or cancels; second call (with confirmation_token) commits the pause. Always include agent_name + app_name so the preview is human-readable.
after · 81% of words in common
Pause an active agent so it stops discovering new leads. The agent's existing leads remain in the list. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms or cancels; second call (with confirmation_token) commits the pause. Always include agent_name + app_name so the preview is human-readable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17enable_agentchangeddescription
before
Enable (activate) a paused or draft agent so it begins discovering leads on its schedule. Two-phase: first call returns a confirmation token + preview; the user reviews and confirms; second call (with confirmation_token) commits.
after · 76% of words in common
Enable (activate) a paused or draft agent so it begins discovering leads on its schedule. Two-phase: first call returns a confirmation token + preview; the user reviews and confirms; second call (with confirmation_token) commits. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17pause_campaignchangeddescription
before
Pause an active campaign. The campaign engine stops scheduling actions. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms or cancels; second call (with confirmation_token) commits the pause. Always include campaign_name + app_name so the preview is human-readable.
after · 77% of words in common
Pause an active campaign. The campaign engine stops scheduling actions. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms or cancels; second call (with confirmation_token) commits the pause. Always include campaign_name + app_name so the preview is human-readable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17create_listchangeddescription
before
Create a new lead list scoped to the request's app. Use this when the user wants to organise leads into a named bucket (e.g. 'Q4 Targets', 'Hot Founders'). The new list starts empty; populate it later via approve_leads with action='approve' and the new list_id, or by editing the list manually.
after · 76% of words in common
Create a new lead list scoped to the request's app. Use this when the user wants to organise leads into a named bucket (e.g. 'Q4 Targets', 'Hot Founders'). The new list starts empty; populate it later via approve_leads with action='approve' and the new list_id, or by editing the list manually. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17draft_messagechangeddescription
before
Generate a personalised draft for one lead. Reads the lead's profile (industry, role, headline, optional reference hook) and returns a single message body. Does NOT send. After the user reviews and approves, call send_message to actually deliver. ALWAYS call get_lead first to inspect reply_state — refuse to draft for opted-out / replied / rejected leads.
after · 78% of words in common
Generate a personalised draft for one lead. Reads the lead's profile (industry, role, headline, optional reference hook) and returns a single message body. Does NOT send. After the user reviews and approves, call send_message to actually deliver. ALWAYS call get_lead first to inspect reply_state — refuse to draft for opted-out / replied / rejected leads. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17enable_campaignchangeddescription
before
Enable (resume) a paused or draft campaign. The campaign engine starts scheduling actions per the campaign's workflow. Two-phase: first call returns a confirmation token + preview; the user reviews and confirms; second call (with confirmation_token) commits. Activation pre-conditions (workflow steps, connected accounts) are validated by the underlying campaign service — the tool surfaces any failure as a structured error.
after · 83% of words in common
Enable (resume) a paused or draft campaign. The campaign engine starts scheduling actions per the campaign's workflow. Two-phase: first call returns a confirmation token + preview; the user reviews and confirms; second call (with confirmation_token) commits. Activation pre-conditions (workflow steps, connected accounts) are validated by the underlying campaign service — the tool surfaces any failure as a structured error. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
2026-10-17update_campaign_workflowchangeddescription
before
Update message content on existing steps in a draft or paused campaign. Prefer one-based message_order, which counts only message-bearing steps and automatically skips invite/follow steps: message_order 1, 2, 3 targets the first, second, and third messages even when raw workflow step 1 is an invite. step_order remains available only when an exact raw workflow order is already known. Provide exactly one selector plus the exact message_template for every item. In message_template use only {FirstName}, {LastName}, and {Company} personalization variables. This is a two-phase confirmed action: first call without confirmation_token returns a confirmation card; approved execution applies the exact stored workflow_messages through the campaign service. It never launches or sends. Always include campaign_name and app_name for a human-readable preview.
after · 92% of words in common
Update message content on existing steps in a draft or paused campaign. Prefer one-based message_order, which counts only message-bearing steps and automatically skips invite/follow steps: message_order 1, 2, 3 targets the first, second, and third messages even when raw workflow step 1 is an invite. step_order remains available only when an exact raw workflow order is already known. Provide exactly one selector plus the exact message_template for every item. In message_template use only {FirstName}, {LastName}, and {Company} personalization variables. This is a two-phase confirmed action: first call without confirmation_token returns a confirmation card; approved execution applies the exact stored workflow_messages through the campaign service. It never launches or sends. Always include campaign_name and app_name for a human-readable preview. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.

Showing the latest 10. Full history: list_changes · trend: quality JSON.

Checks run on what our credential-free, read-only probes observe; tools are never called and no code audit is performed. How it is computed.

Trust 83 / 100

no-repository

provenance55

liveness100

freshness100

hygiene80

+dns-namespace +website +reachable +uptime-100% +updated-4d-ago -no-repository

Derived from observable signals (official registry feed and our own credential-free probes); no code audit performed. How it is computed.

Endpoints

remote	auth	reachable	uptime 30d	p50	protocol	last ok
`https://api.funkel.ai/mcp` streamable-http	none	yes	100%	674 ms	2025-06-18	2026-10-20

Tools (51)

`approve_leads`destructive	Approve or reject one or more discovered leads. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms; second call (with confirmation_token) commits the status flip. action='approve' requires campaign_id (the campaign the leads will be associated with); action='reject' does not. The tool stamps each lead's status + reviewed_at + campaign_id; downstream campaign-engine pipelines ingest approved leads through their normal path. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`configure_agent_source`destructive	Configure X or Hacker News searches for one owned agent and set source listening. First call get_agent_sources. If no source exists, propose 1-10 precise searches from the user's request and saved ICP; do not invent a broad search without showing it. If a source exists, show its current searches and only replace them when the user asks to change them. This tool previews the exact searches, enabled state, and LinkedIn resolution choice, then requires explicit approval. It does not activate a draft or paused agent. For an already configured source whose searches need no change, use set_agent_source_listening instead. X needs a connected X account for collection. Hacker News uses its public search provider. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`control_operator_task`destructive	Pause, resume, cancel, or retry a durable Operator task after exact approval. Supply its current version. Retry also requires one failed idempotent local-read step; completed effects cannot replay. A pause may not stop an already executing provider call. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`create_agent`destructive	Create a new draft agent under the request's scoped app. Two-phase: first call previews every setting; second call after user confirmation creates it. The agent's ICP starts from the app's saved ICP; icp_filters keys you supply override only those keys, every other saved field (job_titles, mandatory_keywords, excluded_companies, company_sizes, ...) is kept as-is. Do not restate the full ICP from memory for a vague request like "set it up" — omit icp_filters entirely to inherit the saved ICP untouched, and only pass the specific keys the user actually asked to change. To deliberately clear a list field, pass it as an empty array. excluded_companies stores exact current employer names. If supplied a LinkedIn company URL, Operator saves its slug name there and retains the URL separately for competitor signals; ask the user to verify the name in the approval. Supply signals only when the user selects exact agent discovery signals; each signal must include its supported platform, category, key, label, enabled state, and config. An explicit empty signals array creates no signals. Without signals, defaults seed from the merged ICP filters. A requested keyword such as product launch can be a keyword_tracking_product_launch signal with config.keyword='product launch'. Ask for missing signal details; never claim a requested setting was saved if this schema cannot represent it. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`create_campaign_draft`destructive	Create a draft campaign in the selected product. First show the exact campaign, agent, destination list, and any agent retargeting for approval. With agent_id and no list_id, use the agent's destination list; if absent, assign the product default list to both. An explicit different list retargets a draft agent only after approval. Never activate, discover, enroll, or send. If workflow_steps are omitted, create the standard 4-step flow. For custom steps ask about type, delay, message mode, template, and final step. In message_template use only {FirstName}, {LastName}, and {Company}. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`create_list`	Create a new lead list scoped to the request's app. Use this when the user wants to organise leads into a named bucket (e.g. 'Q4 Targets', 'Hot Founders'). The new list starts empty; populate it later via approve_leads with action='approve' and the new list_id, or by editing the list manually. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`create_note`destructive	Create a scoped note or reminder after an explicit confirmation of the exact lead, body, and due date. One request_key can be applied only once; a changed intent with the same key fails. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`create_reminder`destructive	Create a scoped note or reminder after an explicit confirmation of the exact lead, body, and due date. One request_key can be applied only once; a changed intent with the same key fails. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`draft_icp_edit`read-only	Stage a proposed ICP edit for one agent. Returns a diff envelope (current vs proposed, fields_changed) so the user can review the change inline before applying it. Does NOT write — to actually apply the edit, call update_icp (confirm-tier) with the same agent_id and proposed fields. Use this when the user describes a change in plain language ('target Series A founders in EU' etc.); never call update_icp directly without first staging here so the user sees the before/after.
`draft_message`	Generate a personalised draft for one lead. Reads the lead's profile (industry, role, headline, optional reference hook) and returns a single message body. Does NOT send. After the user reviews and approves, call send_message to actually deliver. ALWAYS call get_lead first to inspect reply_state — refuse to draft for opted-out / replied / rejected leads. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`enable_agent`destructive	Enable (activate) a paused or draft agent so it begins discovering leads on its schedule. Two-phase: first call returns a confirmation token + preview; the user reviews and confirms; second call (with confirmation_token) commits. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`enable_campaign`destructive	Enable (resume) a paused or draft campaign. The campaign engine starts scheduling actions per the campaign's workflow. Two-phase: first call returns a confirmation token + preview; the user reviews and confirms; second call (with confirmation_token) commits. Activation pre-conditions (workflow steps, connected accounts) are validated by the underlying campaign service — the tool surfaces any failure as a structured error. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`find_leads`read-only	Search leads in the app the request is scoped to. Returns bounded named summaries and IDs, never full rows. Use get_lead to load full detail for a specific id. Default source is 'discovered' (agent-found, awaiting approval); pass source='approved' to search leads already in campaigns. Optional 'search' keyword runs an ILIKE across name/company/headline; optional 'status' narrows by lead status. If results are thin, surface the count and ASK the user before broadening the filters — do not silently relax constraints.
`get_access`read-only	Read current account access for Operator and campaign actions in the selected product. Operator requires active paid or lifetime entitlement; campaign actions also accept active beta. Feature flags are account-wide. This snapshot is not an action authorization or credit quote.
`get_agent_sources`read-only	Read X and Hacker News listening settings for one owned agent, including sources not configured yet. Use list_agents to identify the agent. If configured is false, use configure_agent_source to propose exact searches and enable listening after approval. If configured is true and searches exist, use set_agent_source_listening to toggle without changing searches. A draft or paused agent does not collect until activated. LinkedIn signal groups are separate.
`get_analytics`read-only	Return product- or campaign-scoped activity in a UTC lookback interval. Completed actions and invitations use executed_at; connections and recorded replies use lead event timestamps. These are separate event counts, not one conversion cohort. Default 7 days, max 90.
`get_campaign`read-only	Read one campaign in the selected product. Returns updated_at for exact approved settings changes, plus workflow, sender assignments, schedule, review mode, observed action state, and readiness blockers. Empty templates are not claimed as generated messages. No sending or activation occurs.
`get_conversation_context`read-only	Return bounded conversation context for one lead or conversation in the request scope: lead profile, product/campaign context, last real messages, recent notes, open reminders, and Reply Assistant follow-up suggestions. Use this before meeting prep, reply planning, or questions about what happened with a lead. Read-only; does not draft or send messages.
`get_daily_priorities`read-only	Rank up to four live operational starting points with source IDs and links. Connected senders and insights count across the account; leads, actions, campaigns, and unread messages count within the selected product. Read only.
`get_inbox_queue`read-only	List reply-needed conversations or due follow-ups in the selected product. One predicate determines both the total and page. AI drafts are not counted as sent replies. Limit 50; use offset for later pages.
`get_insight`read-only	Read one insight and its evidence for the selected product, or an explicitly account-scoped insight. Read-only; proposed changes never run here.
`get_lead`read-only	Return full detail for a single lead by id, INCLUDING the derived reply_state and opt_out flags. Always call this before drafting or sending a message — a lead whose reply_state is 'replied', 'rejected', or 'opted_out', or whose opt_out is true, must NEVER be drafted to or messaged. Use source='approved' for leads in active campaigns; default 'discovered' is for agent-found leads awaiting approval.
`get_list`read-only	Read one unarchived list in the selected product and its exact member count. Does not return contacts; use list_list_contacts.
`get_operator_task`read-only	Read one durable Operator task in the selected product with its step states, action states, safe summaries, version, and progress. Does not return executable arguments or raw provider errors.
`get_product_context`read-only	Read the selected product's name, site, description, ICP summary, language, Product Brain availability, and dashboard link. No provider credentials or raw unbounded configuration are returned.
`get_reply_attempt`read-only	Read one reply attempt by its stable request_key in the selected product. Returns actual delivery evidence and next_action. Uncertain delivery requires manual provider review unless an exact outgoing provider message is confirmed. Never retries or sends.
`get_schedule`read-only	List pending, approved, or failed scheduled actions in the selected product, optionally for one campaign. Returns one bounded page and its exact total. Completed actions are not included.
`get_scheduled_action`read-only	Read one action in the selected product: campaign, lead, type, state, scheduled and executed timestamps. Does not return provider metadata or a message body.
`get_sender_health`read-only	Show account-wide connected sender health and separate readiness for campaign assignments in the selected product. An account-wide connection does not prove a campaign has a usable sender.
`import_search_leads_to_list`destructive	Import lead rows from a prior search_external_leads session into a lead list. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms; second call with confirmation_token commits. provider_ids omitted means import every lead in the search session. Supply list_id for an existing list, or list_name to create a new list during the confirmed commit. After a committed result returns list_id, use that exact list_id when creating a campaign from the imported search leads. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`list_agents`read-only	List the specialist agents in the product/app the request is scoped to. Returns a summary (count, brief one-liner, and a list of {id, name, status, type, role, one-line icp_summary}). Internal agent types are mapped to product-facing roles such as lead_finder. Full ICP filters / config are collapsed to a one-liner — call get_agent (later) to load full detail. Use this to ground questions like 'how many specialist agents do I have?' or 'which Lead Finder agents are active?'.
`list_apps`read-only	List the products/apps owned by the request scope's user. Returns a summary (count, brief one-liner, and a list of {id, name, description, url}) with no timestamps or internal flags. Use this for product inventory or explicit product-switch/disambiguation questions. A normal product-scoped request already carries an app_id in Scope; do not call this just to discover which product to use for product-aware actions.
`list_campaigns`read-only	List the campaigns in the app the request is scoped to (excludes archived). Returns a summary (count, brief one-liner, and a list of {id, name, status, agent_id, list_id}). Settings JSON is intentionally omitted; call get_campaign for current workflow, sender, and action detail. Status values: 'draft', 'active', 'paused'.
`list_insights`read-only	List account-scoped and selected-product-scoped open insights. Each item labels its actual scope; account insights are not product-only. Exact total and bounded pages.
`list_list_contacts`read-only	List contacts in one unarchived list in the selected product. Returns exact total and bounded pages. Does not expose email or provider identifiers.
`list_lists`read-only	List unarchived lead lists in the selected product. Returns exact total and bounded pages. A list's cached lead count is not used for the total.
`list_operator_files`read-only	List up to 20 files uploaded to any of your Operator threads in the selected product. Returns file IDs and metadata only, never bytes or extracted text. A file can be used for a same-product LinkedIn reply, not another product or email reply.
`list_operator_tasks`read-only	List durable Operator tasks in the selected product. Returns bounded pages, exact total, task status, version, and links. Read-only.
`list_preferences`read-only	List up to 20 saved Operator preferences for the user and product in the request scope. Values are bounded, labeled customer data, and never override user instructions or system rules. Returns key, value, and source only. No provider credentials or tokens are returned.
`list_reminders`read-only	List open or done lead reminders in the selected product with exact total and bounded pages. Default is open. This read does not complete a reminder.
`pause_agent`destructive	Pause an active agent so it stops discovering new leads. The agent's existing leads remain in the list. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms or cancels; second call (with confirmation_token) commits the pause. Always include agent_name + app_name so the preview is human-readable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`pause_campaign`destructive	Pause an active campaign. The campaign engine stops scheduling actions. Two-phase: first call returns a confirmation token + preview; the user reviews an inline card and confirms or cancels; second call (with confirmation_token) commits the pause. Always include campaign_name + app_name so the preview is human-readable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`prepare_conversation_reply`read-only	Prepare one exact local inbound reply. Returns recipient, sender, channel, inbound revision, final signed body, and optional owned file identity. Never sends. Files are available only for supported connected LinkedIn conversations in the selected product.
`rename_list`destructive	Rename one unarchived list in the selected product after explicit approval. Supply its current name to reject stale changes; include the full new description. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`save_preference`destructive	Save one selected-product Operator preference after explicit approval. The value is bounded JSON data, not a system instruction. Existing keys update. No provider credentials or tokens are accepted. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`search_external_leads`	Search LinkedIn for fresh people leads from the request user's connected account. Use this when the user describes leads to find by words, roles, companies, or market terms, rather than asking to search existing saved leads. Returns a server-owned session_id and up to 10 lead previews. To import the results, call import_search_leads_to_list with that session_id after the user asks to import all or selected leads. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`send_conversation_reply`destructive	Send one exact prepared reply to the latest inbound message after explicit human approval. Provide the complete prepare_conversation_reply result, its attachment_id when present, and a stable request_key. A timeout is uncertain and cannot be retried automatically. LinkedIn files are supported; email files are unavailable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`set_agent_source_listening`destructive	Pause or resume only an existing X or Hacker News source on an agent. First read get_agent_sources. If configured is false or searches are empty, use configure_agent_source instead. The first call shows exact approval; a second user-approved call commits. Saved searches, budgets, LinkedIn resolution and other sources stay unchanged. A draft or paused agent does not collect until activated. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`update_campaign_settings`destructive	Change a campaign's manual review, send window, or sender assignments in the selected product. Read get_campaign first and copy its updated_at as expected_updated_at. Omitted settings remain unchanged. Explicit null sender clears one platform. First call returns an exact approval preview; approved execution saves all changes atomically. It never activates, enrolls, or sends. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`update_campaign_workflow`destructive	Update message content on existing steps in a draft or paused campaign. Prefer one-based message_order, which counts only message-bearing steps and automatically skips invite/follow steps: message_order 1, 2, 3 targets the first, second, and third messages even when raw workflow step 1 is an invite. step_order remains available only when an exact raw workflow order is already known. Provide exactly one selector plus the exact message_template for every item. In message_template use only {FirstName}, {LastName}, and {Company} personalization variables. This is a two-phase confirmed action: first call without confirmation_token returns a confirmation card; approved execution applies the exact stored workflow_messages through the campaign service. It never launches or sends. Always include campaign_name and app_name for a human-readable preview. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.
`update_icp`destructive	Apply an ICP edit (description and/or filters JSON) to one agent. Two-phase: first call returns a confirmation token + preview; the user reviews the diff inline and confirms; second call (with confirmation_token) commits. ALWAYS call draft_icp_edit first so the user sees a before/after; never call update_icp directly without staging a diff. Pre-populate agent_name in the schema args so the preview is human-readable. Over MCP: call this tool once, without confirmation_token. It runs right away and returns the result.

Schemas: list_tools.

Similar MCP servers

Official server.json

{
  "name": "ai.funkel/funkel",
  "title": "Funkel AI",
  "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
  "remotes": [
    {
      "url": "https://api.funkel.ai/mcp",
      "type": "streamable-http"
    }
  ],
  "version": "1.0.0",
  "websiteUrl": "https://funkel.ai",
  "description": "Run LinkedIn and email outreach: find leads, manage campaigns, and reply to prospects."
}