🔧 feat: Initial MCP Support (Tools)

[danny-avila]

Summary

Initial Support for MCP, integrating tools to start, I will then move on to Resources, then Prompts.

Initial tool support is for Agents and Assistants API endpoints.

Note: Agents are recommended due to the ability to provide images from tool results directly to the LLM (Anthropic/OpenAI only).

Relevant issue: #4876

Change Type

• New feature (non-breaking change which adds functionality)

Testing

Tested the following mcpServers config via librechat.yaml file:

# librechat.yaml
mcpServers:
  everything:
      # type: sse # type can optionally be omitted
      url: http://localhost:3001/sse
  puppeteer:
    type: stdio
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-puppeteer"
  filesystem:
    # type: stdio
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-filesystem"
      - /home/danny/LibreChat/
# `iconPath` - unique to LibreChat, defines the tool's display icon shown 
# in the tool selection dialog, opened by "Add Tools" from Agents form
    iconPath: /home/danny/LibreChat/client/public/assets/logo.svg
  mcp-obsidian:
    command: npx
    args:
      - -y
      - "mcp-obsidian"
      - /mnt/c/Users/danny/Documents/obsidian

Checklist

• My code adheres to this project's style guidelines
• I have performed a self-review of my own code
• I have commented in any complex areas of my code
• I have made pertinent documentation changes
• My changes do not introduce new warnings
• I have written tests demonstrating that my changes are effective or that my feature works
• Local unit tests pass with my changes
• Any changes dependent on mine have been merged and published in downstream modules.
• A pull request for updating the documentation has been submitted.

[ThoughtPhotography]

I've been using MCP with Claude Desktop a lot. It's nice to see it in LIbreChat. 👍
Websites that list MCP servers like Glama, and nearly every GitHub repo will have usage examples using .json, so supporting that would be ideal. It's simpler with less chance of messing up the main .yaml file.

claude_desktop_config.json typically contains nothing other than MCP servers, so if the same .json format was supported we could just copy/paste our current MCP config into the LIbreChat dir (renaming it if necessary), or importing .json in the settings (like chat history.)

I attached a bunch of MCP servers converted to .yaml (by Claude). The API keys are redacted, but it might be helpful to have a wider variety of examples.
mcp-example-librechat.txt

Btw, I updated to 0.7.6, but the librechat.yaml is still v1.1.5 (also in the repo.) I tried adding the mcp-servers to v1.1.5., but 18 of 20 failed according to Docker.

Claude .json example.

[epaproditus]

I attempted to set this up on an Ubuntu server at home. While I successfully configured Librechat, the MCP tools did not set up properly. I managed to establish the filesystem MCP server, but when I tried to add another server, like the Todoist MCP server, it failed. I used the deploy-compose.yml file for deployment, and here’s my librechat.yaml configuration. I experimented with various configurations, but all were unsuccessful. I even tried the one mentioned in the earlier testing comment above. I'm not an expert, so I might be overlooking something obvious and would appreciate guidance either way.

mcpServers:
  todoist:
    command: npx
    args:
      - -y
      - "@abhiz123/todoist-mcp-server"
    env:
      TODOIST_API_TOKEN: "REPLACE_WITH_YOUR_API_TOKEN"
  brave-search:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-brave-search"
    env:
      BRAVE_API_KEY: "REPLACE_WITH_YOUR_API_KEY"
  filesystem:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-filesystem"
      - "/Users/abe"
  memory:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-memory"
  github:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-github"
    env:
      GITHUB_PERSONAL_ACCESS_TOKEN: "REPLACE_WITH_YOUR_PERSONAL_ACCESS_TOKEN"
  mcp-installer:
    command: npx
    args:
      - "@anaisbetts/mcp-installer"
  markdown:
    command: npx
    args:
      - -y
      - "mcp-obsidian"
      - "/Users/abe/Library/Mobile Documents/iCloud~md~obsidian/Documents/Vault"
  puppeteer:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-puppeteer"
  airtable:
    command: npx
    args:
      - -y
      - "airtable-mcp-server"
      - "REPLACE_WITH_YOUR_AIRTABLE_API_KEY"
  sequential-thinking:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-sequential-thinking"
  playwright:
    command: npx
    args:
      - -y
      - "@executeautomation/playwright-mcp-server"

[danny-avila]

@ThoughtPhotography appreciate the feedback. We will simply rely on the librechat.yaml file for now. I will see about using JSON from the frontend, though.

[danny-avila]

@epaproditus open a troubleshooting discussion and post relevant logs.

I’m locking discussion here since it’s easily missed and comments here are intended for the actual code.