citedagents.com

Cited Agents is a platform for discovering and running science-backed AI agents. Every agent is grounded in peer-reviewed research, pinned to an exact Git commit, and fully reproducible. ## Workflow 1. **search_agents** — Find agents by keyword, research field, tag, license, or SDK. 2. **get_agent_blueprint** — Get the complete reproducible blueprint for any agent found in step 1. 3. **Reconstruct & run** — Follow the blueprint's spawn_protocol to clone, install, and run the agent locally. ## Understanding Blueprints The blueprint returned by get_agent_blueprint contains everything needed to reconstruct and run the agent: - **spawn_protocol.install**: Exact shell command to clone the repo at the pinned commit and install all dependencies. Run this as-is. - **spawn_protocol.run**: Exact command to execute the agent entrypoint. - **spawn_protocol.stdin / stdout**: Always "json". Pipe JSON input via stdin, read JSON output from stdout. - **spawn_protocol.required_env**: Environment variables that must be set before running (e.g. ANTHROPIC_API_KEY). - **input_schema**: JSON Schema defining the exact input format the agent expects. - **output_schema**: JSON Schema defining the output format the agent returns. - **construction_instructions**: Human-readable step-by-step build instructions. - **examples**: Example inputs you can use for testing. ## To Run an Agent 1. Execute `spawn_protocol.install` in a shell (clones repo at pinned commit + installs deps). 2. Set all environment variables listed in `spawn_protocol.required_env`. 3. Construct a JSON object matching `input_schema`. 4. Run: `echo '{"your":"input"}' | <spawn_protocol.run>` 5. Parse the stdout JSON and validate it against `output_schema`. ## Hosted Execution (Paid Hosted Access) If you are authenticated with an eligible paid organization subscription, you can use the **run_agent** tool to execute hosted agents directly — no local setup, no API keys, no installation needed. The platform handles everything. 1. **search_agents** — Find an agent. 2. **run_agent** — Execute it with JSON input and get results immediately. This requires OAuth authentication (automatic in ChatGPT and Claude MCP clients) and an eligible paid organization subscription with remaining token quota. ## Other Tools - **get_agent** — Get full metadata (owner, org, citations, taxonomy) without the full blueprint. - **list_fields** — Browse OECD research field taxonomy codes for filtering search results. - **get_contributor_kit** — Get the agent.yaml template, validation rules, and submission instructions for building a new Cited Agent. - **validate_agent** — Validate a GitHub repository's agent.yaml before submission.