MCP server for executing code against clinical guidelines and biomedical APIs with support for REST, GraphQL, and SPARQL endpoints. Provides sandboxed V8 isolate execution with auto-staging of large responses.
This server exposes 10 tools, but 7 are internal proxy handlers (__api_proxy, __paginate_proxy, __query_proxy, __stage_proxy, __graphql_proxy, __sparql_proxy, and one duplicate __execute). Only 3 user-facing tools exist: {prefix}_search, {prefix}_execute (REST), {prefix}_execute (GraphQL), {prefix}_execute (SPARQL). The search tool has a detailed description and schema, but the three execute tools are severely under-specified: they accept only a 'code' parameter (JavaScript string) with minimal validation guidance. No output schemas are documented. The internal proxy tools should not be exposed as MCP tools, they are implementation details. Descriptions are present but generic; parameter descriptions lack format constraints, examples, or validation rules. No error handling guidance. The 'code' parameter invites injection attacks and hallucinated syntax without constraints.
Internal proxy handler for REST API calls from within the execute isolate. Routes api.get/api.post calls through the host's HTTP layer with auto-staging of large responses.
Internal proxy handler for GraphQL queries from within the execute isolate. Routes gql.query() calls through the host's GraphQL fetch layer with auto-staging of large responses.
Internal proxy handler for exhaustive pagination through REST API results. Backs api.getAll() calls from within the execute isolate.
Internal proxy handler for SQL queries against staged data in the DO namespace. Backs api.query(data_access_id, sql) calls from within the execute isolate.
Internal proxy handler for SPARQL queries from within the execute isolate. Routes sparql.query/ask/raw calls through the host's SPARQL fetch layer with auto-staging of large responses.
Internal proxy handler for staging data into the DO namespace SQLite. Backs db.stage(data, tableName) calls from within the execute isolate.
Execute JavaScript code against SPARQL endpoints in a sandboxed V8 isolate. Code has access to sparql.query(queryString, opts?), sparql.ask(queryString), sparql.raw(queryString, opts?), and pre-injected ontology prefix maps. Large responses are auto-staged into SQLite. The isolate receives common SPARQL prefixes (rdfs, owl, xsd, skos, dcterms, void, obo, uberon, go, ncbigene, efo, obi, sio, up, ensembl, bgee) and a prefixHeader constant.
Internal proxy tools exposed as MCP tools. __api_proxy, __paginate_proxy, __query_proxy, __stage_proxy, __graphql_proxy, __sparql_proxy are implementation details and should not be registered as user-facing tools. They clutter the tool namespace and confuse LLMs about which tools to call.
Execute tools accept arbitrary JavaScript code with no input validation, format constraints, or injection guards. The 'code' parameter description lacks examples, syntax rules, or error recovery guidance. LLMs will hallucinate invalid syntax or dangerous payloads.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. Responses are opaque, agents cannot know what data structure to expect or how to chain tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2026-07-28+ | v2 |
Execute JavaScript code against GraphQL APIs in a sandboxed V8 isolate. Code has access to gql.query(queryString, variables?) for GraphQL execution, schema.* introspection helpers, and optional REST API support (api.get/api.post). Large responses are auto-staged into SQLite. The isolate receives pre-injected schema helpers and GraphQL proxy functions.
Execute JavaScript code against REST APIs in a sandboxed V8 isolate. Code has access to api.get/api.post wrappers, catalog search helpers, and auto-staging of large responses (>30KB) into SQLite. The isolate receives pre-injected search helpers (searchSpec, listCategories, getEndpoint, describeEndpoint) and api proxy functions. Large responses are auto-staged and can be queried with {prefix}_query_data.
Search the API catalog (catalog mode) or OpenAPI spec (OpenAPI mode). Returns matching endpoints with full parameter docs. Use this to discover API capabilities before calling {prefix}_execute.
Execute tool descriptions are generic and lack actionable guidance. No mention of error handling, timeout behavior, auto-staging thresholds (>30KB), or how to use result.__staged. Agents cannot understand when staging occurs or how to query staged data.
Duplicate tool names: three separate {prefix}_execute tools with identical names but different behaviors (REST, GraphQL, SPARQL). MCP does not support tool name overloading. LLMs cannot distinguish which execute variant to call. Rename to {prefix}_execute_rest, {prefix}_execute_graphql, {prefix}_execute_sparql.