MCP Integration

JamJet has first-class support for the Model Context Protocol (MCP) — the open standard for connecting AI agents to external tools, data sources, and services.

What is MCP?

MCP defines a standard way for AI agents to discover and call external tools. A MCP server exposes tools with typed schemas; a MCP client (JamJet) discovers those tools and calls them.

This means you can connect JamJet to any MCP-compatible tool server — and there are hundreds of community servers for web search, databases, GitHub, Slack, file systems, and more.

Configuring MCP servers

Add servers to jamjet.toml in your project root:

# Local stdio server (most common for dev)
[[mcp.servers]]
name = "brave-search"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-brave-search"]
env = { BRAVE_API_KEY = "${BRAVE_API_KEY}" }

# Local Python server
[[mcp.servers]]
name = "my-tools"
command = "python"
args = ["-m", "my_tools.server"]

# Remote HTTP server (SSE transport)
[[mcp.servers]]
name = "remote-api"
url = "https://tools.example.com/mcp"
headers = { Authorization = "Bearer ${TOOLS_API_KEY}" }

JamJet connects to all configured servers when jamjet dev starts.

Using tools in workflows

YAML

nodes:
  search:
    type: tool
    server: brave-search        # matches `name` in jamjet.toml
    tool: web_search            # tool name from the MCP server
    arguments:
      query: "{{ state.query }}"
      count: 10
    output_key: search_results
    next: summarize

Python SDK

@node
async def search(self, state: State) -> State:
    result = await self.tool(
        server="brave-search",
        tool="web_search",
        arguments={"query": state["query"], "count": 10},
    )
    return {"search_results": result.content}

Discovering available tools

List all tools from connected servers:

jamjet tools list

Server: brave-search
  web_search(query: str, count: int) → list[SearchResult]
  news_search(query: str, freshness: str) → list[NewsItem]

Server: my-tools
  run_query(sql: str) → list[dict]
  get_schema(table: str) → dict

Test a tool directly:

jamjet tools call brave-search web_search --args '{"query": "JamJet runtime", "count": 3}'

Building a MCP server

JamJet can also serve tools over MCP, exposing your workflows as callable tools for other agents.

Add this to jamjet.toml:

[mcp.server]
enabled = true
port = 7701
workflows = ["my-agent", "research-agent"]   # expose these as tools

Any workflow you expose becomes a callable MCP tool. Other agents can discover it, see its input/output schema, and call it — all without knowing it's JamJet underneath.

# Start the MCP server alongside the runtime
jamjet dev --with-mcp-server

Popular MCP servers

Server	Package	What it provides
Brave Search	`@modelcontextprotocol/server-brave-search`	Web + news search
GitHub	`@modelcontextprotocol/server-github`	Repos, issues, PRs
Filesystem	`@modelcontextprotocol/server-filesystem`	Read/write local files
PostgreSQL	`@modelcontextprotocol/server-postgres`	SQL queries
Slack	`@modelcontextprotocol/server-slack`	Messages, channels
Memory	`@modelcontextprotocol/server-memory`	Persistent key-value store

tip: Find the full list of community MCP servers at github.com/modelcontextprotocol/servers.

Tool output

Tool results are available in state under output_key. The shape depends on the tool — check with jamjet tools list for the schema.

For multi-step workflows, intermediate tool results accumulate in state:

nodes:
  search:
    type: tool
    server: brave-search
    tool: web_search
    arguments:
      query: "{{ state.query }}"
    output_key: search_results   # → state.search_results = [...]
    next: docs

  docs:
    type: tool
    server: brave-search
    tool: web_search
    arguments:
      query: "{{ state.query }} documentation"
    output_key: doc_results      # → state.doc_results = [...]
    next: synthesize

Error handling

Tool calls can fail for many reasons — network errors, invalid arguments, rate limits. Configure retry behavior per node:

  search:
    type: tool
    server: brave-search
    tool: web_search
    arguments:
      query: "{{ state.query }}"
    output_key: search_results
    retry:
      max_attempts: 3
      backoff: exponential
      delay_ms: 1000
    next: summarize

If all retries fail, the execution is marked failed and can be manually resumed or retried.

MCP Integration

On this page