MCP client for Discourse sites
Find a file
2025-08-20 13:43:50 +10:00
.cursor/rules coding agent rules 2025-08-18 16:08:58 +10:00
scripts adds filter topics route 2025-08-19 14:46:53 +10:00
src build scripts 2025-08-20 13:25:04 +10:00
.gitignore remove test file 2025-08-20 13:43:13 +10:00
.versionrc.json chore(release): add standard-version and release scripts 2025-08-20 13:41:53 +10:00
AGENTS.md adds filter topics route 2025-08-19 14:46:53 +10:00
CHANGELOG.md chore(release): 0.1.1 2025-08-20 13:43:50 +10:00
LICENSE readme and license 2025-08-18 16:18:07 +10:00
package-lock.json chore(release): 0.1.1 2025-08-20 13:43:50 +10:00
package.json chore(release): 0.1.1 2025-08-20 13:43:50 +10:00
pnpm-lock.yaml chore(release): add standard-version and release scripts 2025-08-20 13:41:53 +10:00
README.md adds filter topics route 2025-08-19 14:46:53 +10:00
tsconfig.json initial commit 2025-08-18 15:58:09 +10:00

Discourse MCP

A Model Context Protocol (MCP) stdio server that exposes Discourse forum capabilities as tools for AI agents.

  • Entry point: src/index.ts → compiled to dist/index.js (binary name: discourse-mcp)
  • SDK: @modelcontextprotocol/sdk
  • Node: >= 18

TL;DR: Configure and run

  • Install and build
pnpm install
pnpm build
  • Run locally (readonly, recommended to start):
node dist/index.js

Then, in your MCP client, call the discourse_select_site tool with { "site": "https://try.discourse.org" } to choose a site.

  • Enable writes (optin, safeguarded):
node dist/index.js --allow_writes --read_only=false --auth_pairs '[{"site":"https://try.discourse.org","api_key":"'$DISCOURSE_API_KEY'","api_username":"system"}]'
  • Use in an MCP client (example: Claude Desktop) — local build:
{
  "mcpServers": {
    "discourse": {
      "command": "node",
      "args": ["/absolute/path/to/dist/index.js"],
      "env": {}
    }
  }
}
  • Use in an MCP client — if installed globally (after publishing or npm -g):
{
  "mcpServers": {
    "discourse": {
      "command": "discourse-mcp",
      "args": []
    }
  }
}

Configuration

The server registers tools under the MCP server name @discourse/mcp. You select the target Discourse site at runtime using the discourse_select_site tool, which validates the site via /about.json.

  • Auth

    • None by default.
    • --auth_pairs '[{"site":"https://example.com","api_key":"...","api_username":"system"}]': Persite API key overrides. You can include multiple entries; the matching entry is used for the selected site.
  • Write safety

    • Writes are disabled by default.
    • The tool discourse.create_post is only registered when all are true:
      • --allow_writes AND not --read_only AND some auth is configured (either default flags or a matching auth_pairs entry).
    • A ~1 req/sec rate limit is enforced for create_post.
  • Flags & defaults

    • --read_only (default: true)
    • --allow_writes (default: false)
    • --timeout_ms <number> (default: 15000)
    • --concurrency <number> (default: 4)
    • --log_level <silent|error|info|debug> (default: info)
    • --tools_mode <auto|discourse_api_only|tool_exec_api> (default: auto)
    • --cache_dir <path> (reserved)
    • --profile <path.json> (see below)
  • Profile file (keep secrets off the command line)

{
  "auth_pairs": [
    { "site": "https://try.discourse.org", "api_key": "<redacted>", "api_username": "system" }
  ],
  "read_only": false,
  "allow_writes": true,
  "log_level": "info",
  "tools_mode": "auto"
}

Run with:

node dist/index.js --profile /absolute/path/to/profile.json

Flags still override values from the profile.

  • Remote Tool Execution API (optional)

    • With tools_mode=auto (default) or tool_exec_api, the server discovers remote tools via GET /ai/tools after you select a site and registers them dynamically. Set --tools_mode=discourse_api_only to disable remote tool discovery.
  • Networking & resilience

    • Retries on 429/5xx with backoff (3 attempts).
    • Lightweight inmemory GET cache for selected endpoints.
  • Privacy

    • Secrets are redacted in logs. Errors are returned as humanreadable messages to MCP clients.

Tools

Builtin tools (always present unless noted):

  • discourse_search
    • Input: { query: string; with_private?: boolean; max_results?: number (150, default 10) }
    • Output: text summary plus a compact footer like:
      { "results": [{ "id": 123, "url": "https://…", "title": "…" }] }
      
  • discourse_read_topic
    • Input: { topic_id: number; post_limit?: number (120, default 5) }
  • discourse_read_post
    • Input: { post_id: number }
  • discourse_list_categories
    • Input: {}
  • discourse_list_tags
    • Input: {}
  • discourse_get_user
    • Input: { username: string }
  • discourse_filter_topics
    • Input: { filter: string; page?: number (default 1); per_page?: number (150) }
    • Query language (succinct): key:value tokens separated by spaces; category/categories (comma = OR, =category = without subcats, - prefix = exclude); tag/tags (comma = OR, + = AND) and tag_group; status:(open|closed|archived|listed|unlisted|public); personal in: (bookmarked|watching|tracking|muted|pinned); dates: created/activity/latest-post-(before|after) with YYYY-MM-DD or relative days N; numeric: likes[-op]-(min|max), posts-(min|max), posters-(min|max), views-(min|max); order: activity|created|latest-post|likes|likes-op|posters|title|views|category with optional -asc; free text terms are matched.
  • discourse_create_post (only when writes enabled; see Write safety)
    • Input: { topic_id: number; raw: string (≤ 30k chars) }

Notes:

  • Outputs are humanreadable first. Where applicable, a compact JSON is embedded in fenced code blocks to ease structured extraction by agents.

Development

  • Requirements: Node >= 18, pnpm.

  • Install / Build / Typecheck / Test

pnpm install
pnpm typecheck
pnpm build
pnpm test
  • Run locally (with source maps)
pnpm build && pnpm dev
  • Project layout

    • Server & CLI: src/index.ts
    • HTTP client: src/http/client.ts
    • Tool registry: src/tools/registry.ts
    • Builtin tools: src/tools/builtin/*
    • Remote tools: src/tools/remote/tool_exec_api.ts
    • Logging/redaction: src/util/logger.ts, src/util/redact.ts
  • Testing notes

    • Tests run with Nodes test runner against compiled artifacts (dist/test/**/*.js). Ensure pnpm build before pnpm test if invoking scripts individually.
  • Publishing (optional)

    • The package exposes a bin named discourse-mcp. After publishing or global install, MCP clients can invoke the binary directly without node dist/index.js.
  • Conventions

    • Focus on textoriented outputs; keep embedded JSON concise.
    • Be careful with write operations; keep them optin and ratelimited.

See AGENTS.md for additional guidance on using this server from agent frameworks.

Examples

  • Readonly session against try.discourse.org:
node dist/index.js --log_level debug
# In client: call discourse_select_site with {"site":"https://try.discourse.org"}
  • Create a post (writes enabled):
node dist/index.js --allow_writes --read_only=false --auth_pairs '[{"site":"https://try.discourse.org","api_key":"'$DISCOURSE_API_KEY'","api_username":"system"}]'

FAQ

  • Why is create_post missing? Youre in readonly mode. Enable writes as described above.
  • Can I disable remote tool discovery? Yes, run with --tools_mode=discourse_api_only.
  • Time outs or rate limits? Increase --timeout_ms, and note builtin retry/backoff on 429/5xx.