Particle Podcast Intelligence

npx add-mcp 'https://mcp.particle.pro'

Particle Pro MCP server

An OAuth-secured MCP server that exposes Particle's podcast, knowledge graph, company, and advertising surfaces as tool calls for AI agents.

The Particle Pro MCP server lives at https://mcp.particle.pro and lets AI agents — Claude Code, Claude Desktop, Cursor, ChatGPT, and any other MCP-aware client — call Particle's podcast, knowledge-graph, company, and advertising tools directly. Setup is a one-time browser approval; everything else is automated by the agent.

Connect your agent

One-click install: Add to Cursor · Add to VS Code — or use the matching manual config below.

```bash Claude Code theme={"dark"} claude mcp add --transport http particle https://mcp.particle.pro ```

// ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
// %APPDATA%\Claude\claude_desktop_config.json (Windows)
//
// Claude Desktop's claude_desktop_config.json doesn't accept a bare `url`.
// Bridge the remote server through `mcp-remote`, which opens a browser
// for OAuth and proxies stdio ↔ Streamable HTTP. Restart Claude Desktop
// after editing.
{
  "mcpServers": {
    "particle": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.particle.pro"]
    }
  }
}

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "particle": {
      "url": "https://mcp.particle.pro"
    }
  }
}

// .vscode/mcp.json
{
  "servers": {
    "particle": {
      "type": "http",
      "url": "https://mcp.particle.pro"
    }
  }
}

# Bridges a stdio MCP client to a remote OAuth-secured Streamable HTTP server.
npx -y mcp-remote https://mcp.particle.pro

Note

Claude Desktop UI alternative: Settings → Connectors → Add custom connector, paste https://mcp.particle.pro, complete OAuth in the browser. Easier than editing JSON and avoids the mcp-remote bridge.

Note

ChatGPT: Settings → Security and login → turn on Developer mode (available to Pro, Plus, Business, Enterprise, and Edu accounts on the web) → chatgpt.com/plugins → + → paste https://mcp.particle.pro as the Server URL with OAuth, and complete the approval in the browser. See the Quickstart for the step-by-step flow.

Any other MCP client: point it at https://mcp.particle.pro over Streamable HTTP and prefer OAuth if the client supports it; stdio-only clients can bridge through mcp-remote (last tab above).

Have an agent do it: paste Read https://api.particle.pro/agents.md and connect me to Particle. into any coding agent. The playbook covers every client above, the restart most clients need before new tools appear, and a first live query.

The first tool call triggers an OAuth flow: your client opens the authorize URL in a browser, you sign in to Particle Pro and pick the project the agent should act as, and your client exchanges the resulting code for a token automatically. See the Quickstart for a step-by-step walk-through and a sanity-check call, or Authentication for the full OAuth 2.1 spec.

What's in the box

40 tools, organized into exposure categories. A bare connection advertises the seven default categories plus the always-on system meta-tools; six more categories, including Radar cards, are opt-in (advertised only when you select them with the ?include= selector). Either way, every public tool is callable by name regardless of what was advertised — see Tool sets & discovery.

Default categories — resolve podcasts/people/companies, search transcripts, list episodes, pull mentions, browse topics, chart rankings, and guests. Opt-in categories — sponsor analytics, publisher profiles, listener ratings, and corpus-wide bias/brand-suitability views. `particle_catalog` and `particle_call` — always on, so a default agent can discover and call the rest of the surface. The selector mechanics and the contract that discovery is free while execution is metered.

Endpoint	https://mcp.particle.pro
Transport	Streamable HTTP
Authentication	OAuth 2.1 (interactive clients) or platform API keys (headless agents) — see Authentication
Tool count	40 across 14 exposure categories (7 default + always-on system, 6 opt-in)
Tool names	Flat particle_* identifiers; every public tool is callable by name
Read/write	Most tools read. particle_alert_create, particle_alert_update, and particle_alert_delete change project state and advertise readOnlyHint: false (delete also destructiveHint: true); particle_call carries the same cautious hints because it can dispatch them

How the tools compose

The tools are not isolated lookups — they expose one connected knowledge graph, and two conventions carry most of the design:

• Lean by default, expand on demand. Most tools return a minimal payload and opt into richer sections via an include array (a company's people, products, and competitors; a person's roles and podcast appearances; an episode's segments, clips, topics, and transcript) or a mode/format switch. A tool does far more than its name implies — check its input schema (or the ↳ expand hints in particle_catalog) before assuming a capability is missing.
• Slugs are edges. Every slug a tool returns — person, company, podcast, episode, publisher, guest — is a valid input to the other tools. Resolve a free-text name once, then traverse: particle_company_resolve("Andreessen Horowitz") → particle_company_get with include: ["people"] → particle_person_get with include: ["podcast_appearances"] → particle_podcast_get_episode with include: ["transcript", "entities"]. Four hops from a company name to every entity discussed in an episode one of its partners appeared on.

Connected clients get the same guidance in-band: the server's initialize instructions state both conventions on every connect, and particle_catalog surfaces each tool's expand options.

Example prompts

For visual comparisons, the optional particle_radar_render_cards tool displays a small selection of results already retrieved by the data tools. The server serves its static MCP Apps UI through resources/read. Rendering is free; clients without UI support receive the same selected evidence and source links in Markdown.

After connecting, try asking your agent:

• "What have podcast hosts said about AI agents replacing SaaS in the last 90 days?"
• "Find the All-In episode where they discussed Anthropic and summarize the segment."
• "Build a brief on Ramp: who runs it, who competes with it, and where its executives have appeared on podcasts."
• "Which podcasts does Notion sponsor, and what do hosts say about it organically?"
• "Who are the trending AI guests this month, and which shows had them?"
• "Everything said about Jensen Huang on podcasts this quarter, organized by theme."

Prompts about a named person, company, or show start from a resolve tool and then traverse the slugs it returns; prompts about a topic start with particle_podcast_search_transcripts. The skill bundles package the same loops with the tools they need.

How it differs from the REST API

The REST API and the MCP server share a single backing process and the same handlers. The differences:

• Shape. Each MCP tool returns one markdown text content block. Pass output_format: "json" to receive the same response as compact JSON in that block instead. If you need typed JSON with a stable schema for programmatic use, hit the underlying REST endpoint linked from each tool page.
• Bundling. Some MCP tools roll up multiple REST calls (e.g. particle_podcast_get_episode can also fetch top entities, segments, clips, topics, and the transcript via its include array). REST endpoints are single-resource.
• Identifiers. MCP tools always accept the agent-facing slug (person_slug, company_slug, podcast_slug, episode_slug, publisher_slug). The REST API accepts the same slugs alongside encoded canonical IDs.
• Auth. Interactive MCP clients use OAuth 2.1 access tokens bound to https://mcp.particle.pro as their aud claim; headless agents may send the same pp_* API keys the REST API accepts. See Authentication.
• Errors. MCP tools surface errors as isError: true results with a human-readable message rather than HTTP problem-details bodies. See Errors.

When to use which surface

• Building an agent (Claude Code, Cursor, Claude Desktop, custom MCP client)? Use the MCP server. The bundled tool responses, slug-first inputs, and markdown rendering are designed for agent loops.
• Building a service (server-to-server, scheduled job, or the backend behind your frontend — API keys must stay server-side)? Use the REST API. Fine-grained control, OpenAPI schema, predictable payloads.
• Both surfaces talk to the same database, the same search ranking, and the same billing meter, so mixing them within one organization is fine.

Discovery

Agents that follow the MCP / OAuth specs discover everything they need from the resource:

• https://mcp.particle.pro/.well-known/oauth-protected-resource — RFC 9728 protected-resource metadata pointing at the AS.
• A 401 from the MCP endpoint carries WWW-Authenticate: Bearer realm="mcp", error="...", resource_metadata="..." so unauthenticated clients can bootstrap.

The Authorization Server's metadata document at https://api.particle.pro/.well-known/oauth-authorization-server carries the rest (token endpoint, registration endpoint, supported scopes, JWKS URI).

Next steps

Step-by-step setup with troubleshooting for each client. OAuth 2.1 end-to-end: discovery, registration, PKCE, refresh, revocation. Every tool, every input, every output. Exposure categories, the include/exclude selectors, and the always-callable contract. How tool errors surface and how agents should react.