Interpro Mcp Server

Overview

Protein families, domains, repeats, and functional sites from InterPro and the member databases it integrates (Pfam, PROSITE, SMART, CDD, PANTHER, and others), read through the keyless EMBL-EBI InterPro API. Search entries, read their full metadata, map a UniProt protein's or a PDB entry's chains to annotations with residue coordinates, walk Pfam, CDD, and PIRSF clans, find proteins that carry a signature, and scan a raw protein sequence. Runs over STDIO, Streamable HTTP, or the public hosted endpoint above, with no credentials to configure; starting new InterProScan jobs needs an operator-set contact email (see Configuration).

Tools

Capability reference

interpro_search_entries _tool

• sourceDatabase (default interpro) plus optional query (1-300 characters), entryType, goTerm (GO:0003677 form), goCategory (P, F, C), and integratedInto (an IPR accession, member databases only); no filter is required, so omitting all of them browses the database
• goTerm and goCategory require sourceDatabase: "interpro"; choose one GO filter per call. Combining them fails before network access because upstream ignores the category when a term is supplied. For member signatures, search InterPro with the GO filter, then read member_databases with interpro_get_entry or search the member database with integratedInto set to the matching InterPro accession
• Cursor-paged, positive integer limit capped at 100 (default 20); larger requests are clamped with a notice. Returns entries, page (total, returned, limit, hasMore, nextCursor), and the applied query; page.limit is the effective page size

interpro_get_entry _tool

• accession required; sourceDatabase is inferred from standard accession prefixes when omitted. A set accession (CL…) fails as entry_not_found and points to interpro_get_clan
• Optional sections selects from description, hierarchy, member_databases, go_terms, literature, cross_references, wikipedia, counters, and overlaps_with. Core metadata and unknown upstream fields remain; omitted sections returns full detail
• Optional positive integer overlapLimit pages complete overlaps_with records; overlapOffset requires it, starts at zero, and advances using overlapsPage.nextOffset. Paging requires overlaps to be retained. Selection and omissions are disclosed; absent or null overlap data stays distinct from an empty list
• Wikipedia thumbnails are omitted unless includeThumbnail: true, and a notice reports what was left out
• entry.descriptionText adds readable text with paragraph, list, and table boundaries alongside the unchanged source entry.description, including its HTML and annotation flags. It is omitted when description is excluded or no readable text is supplied
• A withdrawn record fails as entry_withdrawn with its deletion date and short name

interpro_get_protein_domains _tool

• Canonical six- or ten-character UniProt accession (no isoform suffix); optional residuePosition (1-based, inclusive) keeps every match with a fragment covering it, with all of that match's fragments intact; includeExtraFeatures adds predicted features such as disorder and transmembrane regions
• includeResidueAnnotations: true adds residueAnnotations, the provider's functional rule annotations. A residue filter keeps each whole rule with any fragment covering the position, preserving descriptions, residue letters, locations, and provenance
• Large maps arrive in segments of whole records within a 100,000-byte response budget: matches, requested extra features, then requested residue rules. Follow coverage.nextCursor until hasMore is false. The first segment carries protein metadata; an empty segment cannot establish that nothing is annotated. Residue annotations are omitted until their stage is reached, while {} means that stage returned no selected rules
• Discontinuous fragments, models, and upstream scores are kept as returned; residue_out_of_range and protein_not_found are typed failures

interpro_get_clan _tool

• accession for a Pfam (CL0016), CDD (cl21453), or PIRSF (PIRSF000047) set, with sourceDatabase (pfam, cdd, pirsf) inferred from the prefix; members page by cursor, with positive integer limit capped at 100 (default 20). Larger requests are clamped with a notice; page.limit reports the effective size
• includeRelationships: true adds the relationship graph in slices that fit the response budget; relationshipsPage.nextOffset goes back in as relationshipOffset, which needs includeRelationships set
• includeThumbnail works as on interpro_get_entry; a missing set is clan_not_found

interpro_find_proteins_by_domain _tool

• accession plus optional sourceDatabase; taxonId (NCBI, default 9606) and reviewedOnly (default true; false covers reviewed and unreviewed proteins) are echoed in query
• Optional isFragment selects proteins annotated as fragments (true) or not annotated as fragments (false); omit it to include both. False does not establish sequence completeness. Explicit selections appear in query and must stay unchanged when following cursors
• Optional architectureId accepts protein.ida_accession from interpro_get_protein_domains: a 40-character hexadecimal identifier, normalized lowercase. Equality means the provider's Pfam architecture grouping, not equivalent function. Keep it unchanged when following cursors
• Cursor-paged, positive integer limit capped at 50 (default 20); larger requests are clamped with a notice, and page.limit is the effective size. Ordinary results keep the protein summary, matched entry locations, taxon lineage, and native extra_fields.is_fragment annotation; missing or null status remains unknown. Architecture-filtered rows retain native extra_fields.ida_id when supplied and use detailAvailability to distinguish unavailable entry/taxon details from actual empty arrays
• Optional proteomeId (UP plus nine digits, case-insensitive, canonicalized uppercase; for example protein.proteome from interpro_get_protein_domains) intersects one UniProt proteome with every other selection. taxonId still applies, including its 9606 default, and a proteome returns proteins only when taxonId is its organism or an ancestor. Echoed in query; keep it unchanged when following cursors. Rows then carry the upstream proteomes array, and architecture rows add detailAvailability.proteomes
• No hits is a success with a notice, since the API answers an unknown entry, an unknown taxon, an unknown proteome, and an empty intersection the same way

interpro_get_structure_domains _tool

• pdbId required: a four-character PDB ID such as 1tsr, case-insensitive (accession is accepted as an alias); extended pdb_0000… IDs are not supported. Optional chain restricts the map to one protein chain, case-sensitive (A and a differ); nucleic-acid chains are not listed
• Each match lists its chain records in structures, joined by entry. A record carries locations in chain numbering (start/end, 1-based) beside author numbering (auth_start/auth_end, which can be null or skip), UniProt numbering, and the chain-to-UniProt residue map; protein is null for chains with no UniProt mapping
• InterPro nests at most 20 chain records per match. chainRecordLimitReached flags a match at that limit, which may cover chains not returned; set chain to read one chain completely
• Segments hold whole matches within a 100,000-byte response budget; follow coverage.nextCursor until hasMore is false. The first segment carries structure (metadata, including the protein chains)
• A DNA-only entry succeeds with no matches and a notice; an unknown or obsolete ID fails as structure_not_found, and a chain outside the structure's protein chains as chain_not_found with availableChains

interpro_scan_sequence _tool

• sequence required: one protein sequence as residue letters, optionally after a single FASTA header line (up to 100,000 characters). Case, spaces, and line breaks are ignored; digits, gaps, and stop symbols (*) are rejected, as are empty input and multiple FASTA records
• Sequences already in UniParc return precalculated matches at once (source: "precalculated"). Any other sequence starts one InterProScan 6 job through the EBI Job Dispatcher (source: "interproscan"), which commonly runs about ten minutes, so the usual first result for a novel sequence is a jobId with status submitted, queued, or running
• waitSeconds (0-35, default 30) sets how long a newly started job is awaited before returning its jobId. Repeating a sequence whose job is still running returns the same jobId (tracked per server process); at most 25 jobs are in flight per process, beyond which the call fails as rate_limited with retryAfter
• Without INTERPRO_SCAN_EMAIL set, only precalculated matches are available and any other sequence fails as scan_not_configured (see Configuration). The hosted endpoint does not set it
• Matches are InterProScan records verbatim, sorted by earliest location, 1-based and inclusive, in segments of whole records within a 100,000-byte budget; follow coverage.nextCursor with the same sequence. This tool is not read-only: a call can start a job

interpro_get_scan_result _tool

• jobId required, as returned by interpro_scan_sequence; waitSeconds (0-35, default 30) waits for a pending job, and 0 checks its status once
• Returns status: "finished" with matches, queued or running with no matches (check again later), plus interproVersion and interproscanVersion for a finished job. Large results continue by coverage.nextCursor with the same jobId
• Jobs and results expire seven days after submission; an unknown or expired job fails as scan_not_found, and a job that ended in error as scan_failed, both pointing back to interpro_scan_sequence

interpro_list_reference _tool

• topic required: databases, entry_types, accession_patterns, taxa, or release. The first four are local and make no network call; taxa lists common examples, not every taxon
• release reads the current InterPro version, date, and per-database entry counts, and fails as release_changed if the two upstream responses disagree on the version

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

InterPro-specific:

• 15 database slugs (interpro, pfam, prosite, profile, smart, cdd, panther, cathgene3d, ssf, ncbifam, hamap, prints, pirsf, sfld, antifam) and 11 entry types, with the entry database inferred from standard accession prefixes and conflicting routes rejected locally
• One process-wide request pacer at one upstream request per second for the InterPro API, a 50-second budget per tool call covering pacing, retries, and every request, and a 4 MiB cap on upstream responses
• Sequence scanning checks the keyless Matches API for precalculated results before submitting an InterProScan 6 job, and reuses a running job for a repeated sequence
• Upstream HTTP/1.1 client that handles the API's bodyless 204 responses, which native fetch rejects on Node
• Sparse upstream metadata passes through as null or omitted rather than filled in

Agent-friendly output:

• content[] carries the complete output as compact JSON plus an index of every URL as a link and every image as an image, so clients that read only one surface get the same data
• Opaque cursors bound to their query, with a notice naming the exact next call whenever a result is partial
• Typed failure reasons (entry_not_found, invalid_cursor, upstream_rate_limited, and others) each paired with a recovery hint
• Coordinates describe sequence annotation, not pathogenicity or clinical significance
• Publishes the interpro-guide agent skill (identifier chains, gates, and chain/author/UniProt numbering) over MCP and in plugin installs, the npm package, and the Docker image

Getting started

Connect to the hosted instance, no install:

claude mcp add --transport http interpro-mcp-server https://interpro.caseyjhand.com/mcp

Or run it locally:

claude mcp add interpro-mcp-server -- npx -y @cyanheads/interpro-mcp-server

For other clients, add this to the MCP config file:

{  "mcpServers": {    "interpro-mcp-server": {      "command": "npx",      "args": ["-y", "@cyanheads/interpro-mcp-server"]    }  }}

No API key is needed. For Docker or a source build, see Running the server.

Configuration

The InterPro API is keyless. The server defines one optional variable, and these framework variables are the ones most often set:

See .env.example for the common framework overrides.

Running the server

Local development

Requires Bun v1.4.0 or higher (or Node.js v24+).

• Clone and install:

  sh

  git clone https://github.com/cyanheads/interpro-mcp-server.gitcd interpro-mcp-server && bun installcp .env.example .env   # optional; the server needs no keys

• Build and run:

  sh

  # One-time buildbun run rebuild
  # Run the built serverbun run start:stdio# orbun run start:http

• Run checks and tests:

  sh

  bun run devcheck   # Lint, format, typecheck, securitybun run test       # Vitest test suitebun run lint:mcp   # Validate MCP definitions against spec

Docker

docker run --rm -p 3010:3010 ghcr.io/cyanheads/interpro-mcp-server:latest

The image defaults to HTTP transport, stateless session mode, and logs to /var/log/interpro-mcp-server. To run it over stdio from an MCP client, use "command": "docker" with "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/interpro-mcp-server:latest"]. To build locally, docker build -t interpro-mcp-server .; OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches, with no try/catch in tool logic
• Use ctx.log for request-scoped logging
• Register new tools in the createApp() array in src/index.ts
• Wrap external API calls: validate raw, normalize to the domain type, return the output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheckbun run test

License

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.

Data comes from InterPro and its member databases through the EMBL-EBI API. Licensing varies by member database, so attribute InterPro and the originating database; see InterPro's license page and the EMBL-EBI terms of use.