CodeFlow

You can run the CodeFlow MCP Server to analyze codebases in real time, query the analysis, and integrate with AI assistants or IDEs. It provides real-time insight into code structure, call graphs, and semantic search across Python, TypeScript, and Rust projects, with per-session context and persistent storage for fast reuse.

How to use

You will start the MCP server to enable real-time code analysis and tool-based queries from your MCP-enabled client. The server exposes capabilities for semantic search, call graph generation, function metadata retrieval, and dynamic session context management.

How to install

Prerequisites: Python 3.8 or newer is required. Ensure you have a compatible runtime for any client you plan to use (for example, a Python-based MCP client or an MCP-compatible IDE plugin). You may also need a Node/JavaScript runtime if you use related tooling. Follow the concrete steps below to install and start the MCP server.

# Step 1: Install the MCP server package and CLI tools
# If you have an installation utility, use it as shown
# (Use the exact commands shown in your setup notes to install server tooling)

# Step 2: Start the MCP server with default configuration
code_flow_mcp_server

# Step 3: Start the MCP server with a custom configuration file (optional)
code_flow_mcp_server --config path/to/config.yaml

Configuration and startup details

The MCP server looks for its configuration in codeflow.config.yaml by default. You can customize behavior by providing a configuration file at a specific path when starting the server. The server runs in the background while indexing your codebase, and tools may return empty or partial results during the initial scan.

MCP server connection options

You can connect via two methods if both are exposed in your setup: a remote HTTP endpoint or a local standard input/output (stdio) channel. In this guide, you will use the stdio approach for local development and testing.

Additional notes

The MCP server supports real-time analysis, background maintenance to clean up stale references, semantic search, and graph generation. It also provides a session context to support complex analysis workflows and a suite of tools you can leverage from your MCP client.

Troubleshooting and tips

If you encounter slow indexing or empty replies during the initial scan, check that the server is still indexing and monitor progress with the available status tool in your client. Ensure your configuration file is correctly referenced if you started the server with a custom path.

Security and access

Protect access to your MCP server by restricting network exposure and using secure channels when connecting from clients. Do not expose the server to untrusted networks without appropriate authentication and authorization controls.

Available tools

ping

Test server connectivity and check analysis status, returning the current analysis state and indexed function count.

semantic_search

Search functions semantically using natural language queries and receive analysis status in the response.

get_call_graph

Retrieve the call graph in JSON or Mermaid format, including analysis status.

get_function_metadata

Get detailed metadata for a specific function, including analysis state.

query_entry_points

Fetch identified entry points with pagination and optional full metadata.

generate_mermaid_graph

Create Mermaid diagrams for call graphs to visualize function interactions.

cleanup_stale_references

Manually trigger cleanup of stale file references in the vector store.

update_context

Update the current session context with key-value pairs.

get_context

Retrieve the current session context.

reinforce_memory

Create or reinforce a Cortex memory entry for long-term context.

query_memory

Search Cortex memory with decay-aware ranking.

list_memory

List Cortex memory entries with filters and pagination.

forget_memory

Delete a Cortex memory entry by its identifier.

Resources

Expose top Cortex memories or specific knowledge as MCP resources.