MCP

Both sides of the protocol.

Parallel Research speaks Model Context Protocol in both directions: it consumes tools from external MCP servers and exposes all of its own tools to external MCP clients.

Client

Consume any MCP server

stdio and Streamable HTTP transports, OAuth client-credentials, dynamic tool discovery and automatic reconnect. Tools of connected MCP servers become first-class agent tools — no code changes.

# research.toml
[[mcp.servers]]
name = "postgres"
transport = "stdio"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-postgres"]

[[mcp.servers]]
name = "remote"
transport = "http"
url = "https://mcp.example.com"
oauth = "client-credentials"

Server

Expose the whole toolbox

parallel-research mcp-serve exposes all 49 built-in tools over MCP and actually executes tools/call — any MCP client can drive real web search, contact extraction, memory and browser automation through Parallel.

Browse the tools

parallel-research mcp-serve --transport stdio

# tools/list → web_search, extract_contacts,
#   memory_search, browser_navigate, …
# tools/call → real execution, real results

Transports

stdio for local servers, Streamable HTTP for remote — both client and server side.

OAuth

client-credentials flow handled automatically for authenticated HTTP servers.

Discovery

Tools are discovered dynamically at startup and after reconnects.

Reconnect

Dropped stdio/HTTP servers are reconnected without restarting the agent.