Mission Squad MCP Server

The Mission Squad MCP server (@missionsquad/mcp-msq) exposes the MissionSquad API to AI agents as Model Context Protocol tools. With it, an agent can manage providers, models, agents, workflows, factories, schedules, collections, vector stores, and files — and run chat completions and embeddings — using a single MissionSquad API key.

Hosted endpoint: https://mcp.missionsquad.ai — use the remote server without installing anything.
Local: run it over stdio with npx -y @missionsquad/mcp-msq.
Auth: your MissionSquad API key (msq-...). The server forwards it to the API as the x-api-key header.
Tools: 76 tools, all prefixed msq_, returning pretty-printed JSON.

Every tool call ultimately hits the MissionSquad REST API documented in the API reference — the MCP server is a thin, typed wrapper, so the request/response shapes match the corresponding REST endpoints.

Connecting

Hosted (remote)

Point your MCP client at https://mcp.missionsquad.ai and authenticate with your MissionSquad API key. The server resolves the key from a per-call hidden argument named apiKey — a FastMCP feature where arguments not declared in a tool's schema are passed through to the server but never shown to the model. In practice you supply apiKey alongside each tools/call request; clients that support hidden/passthrough arguments inject it for you. You can also pass an optional baseUrl hidden argument to target a self-hosted MissionSquad API.

jsonc

// A tools/call request carries the hidden apiKey (and optional baseUrl) in `arguments`.
// These keys are not part of any tool's published schema, so the model never sees them.
{
  "method": "tools/call",
  "params": {
    "name": "msq_list_models",
    "arguments": {
      "apiKey": "msq-************************",
      "baseUrl": "https://agents.missionsquad.ai/v1"   // optional
    }
  }
}

The hosted server is in active rollout. If https://mcp.missionsquad.ai is not yet reachable for your account, use the local stdio install below — it is functionally identical.

Local (stdio)

Run the published package directly. It speaks the MCP stdio transport, which every MCP client supports.

jsonc

{
  "mcpServers": {
    "mission-squad": {
      "command": "npx",
      "args": ["-y", "@missionsquad/mcp-msq"],
      "env": {
        "MSQ_API_KEY": "msq-************************",
        "MSQ_BASE_URL": "https://agents.missionsquad.ai/v1"
      }
    }
  }
}

Configuration

Setting	Where	Purpose
API key	env `MSQ_API_KEY`, or hidden arg `apiKey`	Authenticates to the MissionSquad API (sent as `x-api-key`). The hidden arg takes precedence over the env var. Required.
Base URL	env `MSQ_BASE_URL`, or hidden arg `baseUrl`	Upstream MissionSquad API base (default `https://agents.missionsquad.ai/v1`). Must end in `/v1`.
`MSQ_HTTP_TIMEOUT_MS`	env	Per-request timeout for non-streaming calls (default 120000).
`MSQ_DEFAULT_FILE_CONTENT_MAX_BYTES`	env	Default cap on bytes returned by `msq_get_file_content` (default 1048576).
`LOG_LEVEL`	env	`debug` / `info` / `warn` / `error` (logs go to stderr).

If no API key resolves, every tool fails with: "MissionSquad API key is required. Provide hidden argument 'apiKey' or set MSQ_API_KEY."

Tool reference

All tools are prefixed msq_. Tools marked (no params) take only the hidden apiKey/baseUrl. Parameters below are the declared input schema; see the linked REST page for full response shapes.

Models & providers

Tool	Params	REST
`msq_list_models`	(no params)	GET /v1/models
`msq_get_model_map`	(no params)	GET /v1/modelmap
`msq_list_providers`	(no params)	GET /v1/core/providers
`msq_add_provider`	`providerKey`, `apiKey?`, `url?`	POST /v1/core/add/provider
`msq_delete_provider`	`providerKey`	POST /v1/core/delete/provider
`msq_discover_provider_models`	`providerKey`, `url?`, `apiKey?`	POST /v1/core/models
`msq_add_model`	`name`, `description`, `providerKey`, `model`, `testResponse?`, `getAllApiModels?`, `extractEmbeddingModels?`	POST /v1/core/add/model
`msq_delete_model`	`modelId`	POST /v1/core/delete/model

Note: in msq_add_provider/msq_discover_provider_models, the apiKey parameter is the provider's key sent in the request body — distinct from the hidden apiKey used to authenticate to MissionSquad.

Chat & embeddings

Tool	Params	REST
`msq_chat_completions`	`model`, `messages[]`, `temperature?`, `max_tokens?`, `top_p?`, `n?`, `stop?`, `tools?`, `tool_choice?`, `stream?`, `xClientId?`, `xSessionId?` (+ passthrough)	POST /v1/chat/completions
`msq_embeddings`	`model`, `input` (string or string[])	POST /v1/embeddings

msq_chat_completions always streams upstream and reassembles the full response; xClientId/ xSessionId are sent as the x-client-id/x-session-id headers.

Agents

Tool	Params	REST
`msq_list_agents`	(no params)	GET /v1/core/agents
`msq_add_agent`	`name`, `description`, `systemPrompt?` or `systemPromptId?`, `model`, `overwrite?`, `addToday?`, `timezoneOffset?`, `tools?`, `selectedFunctions?`, `modelOptions?`	POST /v1/core/add/agent
`msq_update_agent`	`name`, `description?`, `systemPrompt?`/`systemPromptId?`, `model?`, `addToday?`, `timezoneOffset?`, `tools?`, `selectedFunctions?`, `modelOptions?`, `combineSystemPrompts?`, `convertSystemPrompt?`	PUT /v1/core/update/agent
`msq_delete_agent`	`name`	POST /v1/core/delete/agent
`msq_publish_agent`	`agentId`, `agentName`	POST /v1/core/agents/publish
`msq_generate_prompt`	`model`, `messages[]`, `name?`, `description?`, `type?`, `modelOptions?`	POST /v1/core/generate/prompt

msq_add_agent exactly one of systemPrompt or systemPromptId is required; for long prompts call msq_generate_prompt first and pass the returned systemPromptId. msq_add_agent also auto-publishes the agent.

Workflows

Tool	Params	REST
`msq_list_workflows`	(no params)	GET /v1/core/workflows
`msq_get_workflow`	`id`	GET /v1/core/workflows (filtered)
`msq_create_workflow`	`id?`, `name?`, `mainAgentRef?`, `mainAgentId?`, `mainPrompt?`, `dataPayload?`, `concurrency?`, `delimiter?`, `failureMessage?`, `failureInstruction?`	POST /v1/core/workflows
`msq_update_workflow`	`id` + workflow fields	PUT /v1/core/workflows/:id
`msq_run_workflow`	`workflowId`, `dataPayload?`	POST /v1/core/workflow-runs
`msq_get_workflow_run_status`	`runId`	GET /v1/core/workflow-runs/:runId
`msq_get_workflow_result`	`runId`	GET /v1/core/workflow-runs/:runId/hydrated

Factories

Tool	Params	REST
`msq_list_factories`	(no params)	GET /v1/core/factories
`msq_get_factory`	`id`	GET /v1/core/factories/:id
`msq_create_factory`	`id?`, `name`, `description?`, `steps[]`, `continuous?`, `limitTotalInvocations?`, `maxTotalInvocations?`	POST /v1/core/factories
`msq_update_factory`	`id` + factory fields (partial)	PUT /v1/core/factories/:id
`msq_delete_factory`	`id`	DELETE /v1/core/factories/:id
`msq_list_factory_runs`	`factoryId`, `limit?`, `offset?`	GET /v1/core/factories/:id/runs
`msq_run_factory`	`factoryConfigId`, `initialCarryPayload?`	POST /v1/core/factory-runs
`msq_get_factory_run_status`	`runId`	GET /v1/core/factory-runs/:runId
`msq_get_factory_result`	`runId`	GET /v1/core/factory-runs/:runId
`msq_list_factory_run_steps`	`runId`, `limit?`, `offset?`	GET /v1/core/factory-runs/:runId/steps
`msq_get_factory_run_step`	`runId`, `stepRunId`	GET …/steps/:stepRunId
`msq_get_factory_run_step_hydrated`	`runId`, `stepRunId`	GET …/steps/:stepRunId/hydrated
`msq_pause_factory_run`	`runId`	POST …/pause
`msq_resume_factory_run`	`runId`	POST …/resume
`msq_cancel_factory_run`	`runId`	POST …/cancel

Factory schedules

Tool	Params
`msq_list_factory_schedules`	(no params)
`msq_create_factory_schedule`	`factoryConfigId`, `timesToRun[]`, `label?`, `startDate?`, `repeatInterval?`, `daysOfWeek?`, `dayOfMonth?`, `status?`, `initialCarryPayload?`
`msq_update_factory_schedule`	`id` + schedule fields (partial)
`msq_delete_factory_schedule`	`id`
`msq_toggle_factory_schedule`	`id`

Config, tools & servers

Tool	Params	REST
`msq_get_core_config`	(no params)	GET /v1/core/config
`msq_get_core_config_summary`	(no params)	GET /v1/core/config (summarized + redacted)
`msq_scrape_url`	`url`	POST /v1/core/scrape-url
`msq_list_tools`	(no params)	GET /v1/core/tools
`msq_list_tool_functions`	(no params)	GET /v1/core/tools (summarized)
`msq_list_servers`	(no params)	GET /v1/core/servers
`msq_list_server_tools`	`serverName`	GET /v1/mcp/servers/:name/tools
`msq_get_user_settings`	(no params)	GET /v1/core/user/settings

Local collections

Tool	Params
`msq_list_core_collections`	(no params)
`msq_search_core_collection`	`collectionName`, `query`, `embeddingModelName`, `topK?`
`msq_get_core_collection_diagnostics`	`collectionName`
`msq_recover_core_collection`	`collectionName`, `strategy?`, `force?`

Vector stores & files

Tool	Params	REST
`msq_list_vector_stores`	(no params)	GET /v1/vector_stores
`msq_create_vector_store`	`name`, `file_ids?`, `chunking_strategy?`, `metadata?`, `embeddingModelName?`, `enhancePDF?`, `sseSessionId?`, `batchSize?`	POST /v1/vector_stores
`msq_get_vector_store`	`vectorStoreId`	GET /v1/vector_stores/:id
`msq_delete_vector_store`	`vectorStoreId`	DELETE /v1/vector_stores/:id
`msq_list_vector_store_files`	`vectorStoreId`	GET /v1/vector_stores/:id/files
`msq_add_vector_store_file`	`vectorStoreId`, `file_id`, `chunking_strategy?`, `enhancePDF?`	POST /v1/vector_stores/:id/files
`msq_get_vector_store_file`	`vectorStoreId`, `fileId`	GET …/files/:fileId
`msq_get_vector_store_file_details`	`vectorStoreId`	GET …/file-details
`msq_cancel_vector_store_session`	`sessionId`	POST /v1/vector_stores/cancel
`msq_list_files`	(no params)	GET /v1/files
`msq_upload_file`	`filePath`, `purpose`, `relativePath?`, `collectionName?`, `filename?`	POST /v1/files (multipart)
`msq_get_file`	`fileId`	GET /v1/files/:id
`msq_delete_file`	`fileId`	DELETE /v1/files/:id
`msq_get_file_content`	`fileId`, `maxBytes?`	GET /v1/files/:id/content
`msq_list_user_collections`	(no params)	GET /v1/user-collections

msq_upload_file reads filePath from the server's local filesystem. On the hosted server you cannot reference your own local files through this tool; upload via the Files API and reference the resulting fileId instead.

Scheduled runs

Tool	Params
`msq_list_scheduled_runs`	(no params)
`msq_create_scheduled_run`	`agentName`, `prompt`, `startDate`, `timesToRun[]`, `repeatInterval`, `daysOfWeek?`, `dayOfMonth?`, `status?`, `sendEmail?`, `deliveryMethod?`, `slackWebhookUrl?`, `slackMetadata?`
`msq_update_scheduled_run`	`id` + scheduled-run fields (partial)
`msq_delete_scheduled_run`	`id`
`msq_toggle_scheduled_run`	`id`
`msq_get_scheduled_run_results`	`id`

Flagship tool schemas (verbatim)

msq_create_workflow:

{
  id?: string                  // generated when omitted
  name?: string                // default "Untitled Workflow"
  mainAgentRef?: string | null // "agent/<agentId>" or "shared/<ownerUsername>/<slug>"; null clears
  mainAgentId?: string | null  // legacy; translated to "agent/<id>"
  mainPrompt?: string          // contains {{data}} templates and <selector|#|dataKey> helpers
  dataPayload?: string         // JSON string; must be valid JSON if provided
  concurrency?: number         // max concurrent helper executions
  delimiter?: string           // default "|#|"
  failureMessage?: string
  failureInstruction?: string
}

msq_run_factory:

{
  factoryConfigId: string
  initialCarryPayload?: string  // defaults to "" on the backend
}

Mission Squad MCP Server ​

Connecting ​

Hosted (remote) ​

Local (stdio) ​

Configuration ​

Tool reference ​

Models & providers ​

Chat & embeddings ​

Agents ​

Workflows ​

Factories ​

Factory schedules ​

Config, tools & servers ​

Local collections ​

Vector stores & files ​

Scheduled runs ​

Flagship tool schemas (verbatim) ​

See also ​