AI Open Source Intelligence

What AI Open Source Intelligence MCP server does

The AI Open Source Intelligence MCP server connects an MCP-compatible host to public AI Open Source Radar data. It is designed for research tasks involving open-source AI projects, including finding candidates from explicit requirements, checking facts about a named project, comparing options, and planning a possible stack.

Results distinguish source-backed observations from host-model recommendations, unresolved information, and identified risks. This separation is useful when a project appears to match a requirement but the available evidence is incomplete. License results are technical evidence only and are not legal advice.

The product has one active Skill, named ai-open-source-intelligence. That Skill routes research requests to the available MCP tools rather than asking users to select separate research, comparison, or stack-planning Skills.

How it works

The host model selects and calls the read-only tools, then synthesizes the response. The server provides Radar data and evidence; it does not perform a second AI Workstation server-side model call in the standard workflow. Requirement-based selection uses the public Radar selector with model execution disabled.

Formal requirements are represented as typed constraints, such as a required web UI. Matches are checked against project-detail evidence, and unresolved hard requirements remain blockers instead of being silently relaxed. The response also reports observation time and evidence-source freshness. Official publisher links are kept in a separate data.official_resources field so attribution does not alter research conclusions.

Tools and capabilities

The nine tools are:

• search_ai_projects: find and verify projects against explicit requirements.
• get_project_facts: retrieve current public facts for one project.
• get_license_evidence: inspect observed license evidence from available sources.
• compare_ai_projects: compare two to five projects for one decision context.
• find_alternatives: identify candidates while retaining stated constraints.
• compose_ai_stack: propose a candidate stack and expose unverified compatibility.
• get_radar_overview: inspect rankings, collections, categories, scenarios, and filters.
• browse_radar_projects: browse or filter Radar projects by supported groupings.
• browse_radar_skills: browse the Radar Skills library or open a Skill by ID.

The AI Open Source Intelligence MCP server does not execute or install third-party repository code. It can surface compatibility unknowns, but it does not claim cross-project compatibility without evidence or a controlled test.

Setup and configuration

The hosted endpoint is https://mcp.aiworkstation.cn/mcp. Hosted access is anonymous, read-only, data-only, and requires no OAuth. ChatGPT web users can register it as a No Authentication developer-mode app while the public directory listing is pending. Codex and ChatGPT desktop Codex users can install the version-pinned repository Plugin.

For local Python use, install the MCP extra at version 0.3.3:

python -m pip install "aiworkstation-open-source-intelligence[mcp]==0.3.3"

Run local development with osi-mcp. OSI_PROVIDER=mock uses offline fixtures; OSI_PROVIDER=http uses live public Radar data and can be paired with AIWORKSTATION_RADAR_BASE_URL=https://aiworkstation.cn.

Limitations and notes

All nine tools are read-only. The current hosted release has no Premium model tool, checkout flow, OAuth mode, or server-side AI Workstation model execution. Setting hosted access to OAuth fails closed.

License labels are not treated as verified licenses unless supported by a direct LICENSE-family source. Missing licenses do not imply permission. The server also does not replace legal review, execute security tests, install software, or validate compatibility through cross-project runtime testing.

Hosted traffic is subject to per-IP request and connection limits, including a 256 KB maximum MCP request body. The public repository is Apache-2.0 licensed, but that license does not grant rights to private databases, unpublished datasets, credentials, infrastructure, or trademarks.